[Date Index] [Thread Index] [Author Index]

Re: Problems about NSolve

• To: mathgroup at smc.vnet.net
• Subject: [mg37322] Re: [mg37303] Problems about NSolve
• From: Daniel Lichtblau <danl at wolfram.com>
• Date: Wed, 23 Oct 2002 02:57:04 -0400 (EDT)
• References: <200210220847.EAA27537@smc.vnet.net>
• Sender: owner-wri-mathgroup at wolfram.com

Gabriele wrote:
> 
> I want solve a non linear system of 4 algebric equations with the command
> NSolve.
> 
>  I get the error messages:
> 
> NSolve::zerosol  Scale of possible solution value -8.656393839455443`*^-7
> for variable A4 is too small. Must set this value to zero in order to avoid
> generation of spurious solutions. This may cause the loss of some solutions.
> 
> and:
> 
> NSolve::sfail: Subsystem could not be solved forA3 at value
> 0.00002176757227083404`. The likely cause is failure to detect zero due to
> low precision. The likely effect is the loss of one or more solutions.
> Increasing WorkingPrecision might prevent some solutions from being lost.
> 
> Where can I find informations about these?
> 
> Thank you very much and sorry for my english.


The messages indicate that the NSolve algorithm is running into a
problem with ill conditioning. This may be due to multiplicity of the
solution set of perhaps just multiplicity (or near multiplicity) of one
or more coordinate values in the solution set.

The latter type of problem has been addressed in version 4.2. Was this
run in version 4.2 of Mathematica? If not, you might get better results
there. Hard to say for certain since you did not post the actual
equations.

Another possibility might be to set WorkingPrecision high, say to 400,
and see what happens.


Daniel Lichtblau
Wolfram Research

• References:
  • Problems about NSolve
    • From: "Gabriele" <gabriele.bb@tin.it>